Multi-period vehicle assignment with stochastic load availability

This work investigates optimization techniques for a vehicle-load assignment problem.
A company owning a limited fleet of vehicles wants to maximize its operational profit over an infinite horizon divided into periods. The profit stems from revenues for transporting full truckloads and costs derived from waiting idle and moving unladen.
The stochastic component of the problem arises from projections on the realization of each transportation order, i.e. load. The methodology is based on optimizing decisions for deterministic scenarios. Several policies are generated in this way, from simple heuristics to more complex approaches, such as consensus and restricted expectation algorithms, up to policies derived from network flow models formulated over subtrees of scenarios. Myopic and a-posteriori deterministic
optimizations models are used to compute bounds allowing for performance evaluation.
Tests are performed on various instances featuring different number of loads, graph sizes, sparsity, and probability distributions. Performances are compared statistically over paired samples. The robustness of various policies with respect to erroneous evaluations of the probability distributions is also analyzed.